A new mini PC aims to bring large language model inference to desktop users without requiring expensive server hardware. Mechrevo is preparing the Master Road X10 to address the memory bottlenecks that typically limit local AI performance. This device targets researchers and hobbyists who need to run complex models but lack access to data center resources.
Proprietary storage technology offloads large AI models from DRAM to NAND flash
The system relies on Nvidia's RTX Spark platform to combine processing power with high-speed memory. It features a 20-core Nvidia Grace CPU paired with a Blackwell GPU that includes up to 6,144 CUDA cores. This architecture allows the machine to handle heavy computational loads in a compact form factor.
Specifications
- CPU: 20-core Nvidia Grace CPU
- GPU: Blackwell GPU with up to 6,144 CUDA cores
- Memory: Up to 128 GB LPDDR5X unified memory
- Storage: Proprietary AI SSD
Memory capacity reaches up to 128 GB of LPDDR5X unified memory, which helps reduce latency between the processor and graphics unit. Mechrevo also introduced a proprietary AI SSD designed to manage large model weights more efficiently. This storage technology can offload portions of AI models from DRAM to NAND flash when memory runs low.
Mechrevo claims this AI SSD configuration increases supported model workloads by up to 5.7 times compared to standard setups. The company also states the technology can accelerate the prefill stage of model inference by as much as 95 percent. These improvements allow the device to run models with 80 billion, 120 billion, and 122 billion parameters while maintaining responsive performance.
Exact specifications for the SSD capacity, final pricing, and release date remain unknown at this time. Mechrevo has not yet published detailed benchmarks or test conditions to substantiate these performance claims. Buyers should wait for official confirmation before making purchasing decisions based on these preliminary figures.
Source: NotebookCheck
Discussion
0 comments